Figma Moves to Keep AI Agents Inside Brand Rules

Figma announced a set of platform updates at its first Config event in India to define how AI agents work on its platform. The announcements focus on keeping agent output within brand rules, expanding what agents can do on the canvas, and connecting Figma to external agent tools. TechCrunch

The main release is Figma Specs, an editor based on Markdown, a simple plain-text format, for defining a design language of colors, components and rules. A spec can live as a standalone file in Figma or alongside work on the canvas. Users can link directly to canvas elements from a spec and embed FigJam diagrams, code layers and designs inside the document.

Teams can bring design libraries, code libraries and guidelines into Specs to manage context for the design system in one place. Agents then refer to that central source to stay within the rules. The stated goal is for Figma's own agents and third-party agents to follow a brand's design rules more closely while creating with AI.

Figma also made its agent on the canvas available to more users. The agent in Figma Design can generate and remix designs, handle routine tasks and give design feedback. Earlier work described the agent creating design layers to clarify intent across flows, states, copy and structure, with skills, or added instructions, used to guide agents with context.

Code layers are now in public beta. Figma had announced the feature in private beta in June. Code layers add working code to designs on the canvas for more interactive prototypes. Users can add a code layer from the toolbar, create one from an existing frame, or ask the Figma agent to generate one. Users can ask the agent to turn a design into code or change part of the code to iterate on a prototype.

Interoperability is the other thread. Figma is supporting over 20 new tools through its MCP, a standard connection that lets AI tools work with Figma, including Open Code, Dia, Grok Bot and Sarvam. That extends a canvas that Figma says AI agents can design directly on, a position it laid out in March when it opened the canvas to agents.

Figma is also integrating Payload with its Figma Make product for generating dynamic websites and managing content directly. Figma acquired Payload, an open-source headless CMS, or system that stores content separately from page presentation, last year. Figma said Payload's downloads have increased sixfold in the last 12 months.

The choice of venue reflects usage. Figma said India is its biggest market in terms of users outside the United States. It said its platform works with 40% of India's top 100 companies listed on the Bombay Stock Exchange.

Co-founder and CEO Dylan Field has framed the current phase as "AI has lowered the floor but has not raised the ceiling, and that designers, creatives and builders will raise the ceiling." The agent is exiting beta with improved latency, or response speed, and the ability to search for Figma files in chat, while shader generation in Figma Design remains in open beta. Config 2026 launches included code layers, Figma Motion, shaders, generative plugins and Weave tools on the canvas.

The broader context here is control versus generation. Most design tools can now generate screens quickly. The familiar failure is plausible layouts that drift from tokens, or saved design values, components and production limits. Specs looks like an attempt to keep the design system readable by machines where the work happens, rather than as separate documentation agents ignore. If libraries, code and guidelines can be checked at generation time, review becomes comparing against a spec instead of marking up pixels.

In my view, two practical tests are worth watching. First, whether Specs stay current when code libraries change, or become another file to maintain. Second, whether wider MCP support leads to reliable round-trips between Figma and coding tools and bots without losing structure. The optimistic case is clear. A shared spec plus working code layers and a content path through Make shortens the distance from idea to working, content-driven prototype. That is useful leverage for product teams, if the constraints hold under real production pressure.